Immetrica
Capping of tuning-without-viewing

A key problem in RPD (return path data) systems is their inability to determine when the monitor is on. Various methods are used to estimate the probability that part of a long period of viewing of the same channel is tuning-without-viewing. Unfortunately none of the large global TAM providers that also undertake RPD processing publish their algorithms in even general terms, instead treating them as industrial secrets and making comparison between them almost impossible. We are successfully removing most of this RPD-specific tuning-without-viewing using transparent algorithms and are working on improved and conceptually different algorithms.
The HDMI cabling standard for HD and UHD (4k) sets offers information on the monitor’s power state. This is a long-term solution, though, because SD sets are still the majority of almost any operator’s universe (installed base) and likely to remain so for years. Also no data collection system currently can query HDMI, and someone would have to fund the development of this feature. Finally, a certain percentage of viewers will always avoid HDMI, either because they use irrelevant devices (computers or smartphones via streaming to the monitor) or because they do not wish to be subject to HDMI’s copyright enforcement. But in combination with VOD and OTT, this might become a solution in most cases—years from now—to tuning-without-viewing that is an artifact of RPD.
However, tuning-without-viewing for other reasons (TV on with no-one in the room, someone in the room but not paying attention, the viewer fell asleep with the TV on) may well remain a poorly tractable problem as it does in TAM. Although (very expensive) meters exist that monitor a human presence, by means of a nonrecording camera, there has been tremendous resistance by sample-member households to their use, and they were never deployed in production. RPD cannot use such devices because it relies on software deployed in standard STBs to reduce costs. Thus, the only possibly promising approaches are collection of ancillary interaction (such as volume changes, already supported by some data collection systems) and/or statistical analysis. We are pursuing both those approaches.